EEG Motor Imagery Classification using Frequency-Domain and Spatial Filtering Methods: A Comparative Study

This technical report presents a comparative evaluation of bandpower and CSP-based feature extraction for EEG motor imagery classification. Using PhysioNet EEGBCI data from three subjects, Logistic Regression with bandpower features achieved 67.4% accuracy (F1: 0.683), outperforming CSP features (61.5%, F1: 0.660) in a 5-fold cross-validation setup. Full code and methodologyare documented for reproducibility.